Skip to content
Theme
OnDevice Tools

AIChatbots & Assistants

AI in agriculture needs local field data to avoid crop damage

TechRadar reports that general-purpose AI chatbots can give damaging farming advice without local data, and researchers say purpose-built agricultural AI is needed.

Illustration of a farmer checking an AI chatbot beside damaged sesame seedlings

Key takeaways

  • A chemical mix suggested by a chatbot killed all sesame seedlings on 25 acres in China, TechRadar reports.
  • A Nature Food study found generative AI gave unclear or unsafe advice on pests and herbicides.
  • About 17 percent of farmers use generative AI, according to a McKinsey survey cited by TechRadar.
  • Purpose-built agricultural AI needs local field data and expert limits to reduce harmful advice.
In this article
  1. AI chatbot advice kills sesame seedlings
  2. Study finds unsafe chatbot advice
  3. Why farmers rely on chatbots
  4. What field-specific AI would need
  5. Specialised assistants and barriers

AI chatbot advice kills sesame seedlings

According to TechRadar, a farmer in Chuzhou, China, used a general-purpose AI chatbot in July for advice on weeds and pests. The case shows why AI in agriculture can fail without local field data. The chatbot suggested a chemical mixture that was not suitable for sesame.

The farmer sprayed the mixture across 25 acres. All seedlings were dead by the next day. When asked what went wrong, the chatbot admitted errors. TechRadar reports a loss of about 150,000 yuan, or roughly 18,000 euros.

The chatbot had chosen a mix for broadleaf weeds. Sesame is also a broadleaf plant, so the option was wrong from the start. It also advised spraying the whole field rather than using a targeted approach.

Study finds unsafe chatbot advice

Illustration: ChatGPT logo, AI, satellite, market trend and the labels AI in Agriculture, Arm, Generative AI

The TechRadar article also cites a study published in Nature Food. Researchers from the United States, the United Kingdom, Kenya, Nigeria and Colombia tested whether generative AI could replace human farm advisers. Generative AI is software that creates text and recommendations.

They asked chatbots how to treat fall armyworm, a pest that damages corn. OpenAI models GPT-3.5 and GPT-4 gave unclear advice about pesticides, which are chemicals used to control pests. In a separate test, ChatGPT named herbicides, or weed-killing chemicals, for cassava farmers in Nigeria but gave the wrong timing. TechRadar says the advice could have destroyed crops in the most important cassava producer in Africa.

TechRadar says such tools can sound professional while giving poor guidance. The article points to the need for agricultural data and expert limits.

Why farmers rely on chatbots

TechRadar says many farmers turn to chatbots because professional advice is costly and scarce. Labour, land and equipment costs are rising. Trained agricultural extension workers, who advise farmers, are in short supply.

Remote fields in South America and Africa often receive little support. Farmers may not be able to travel hundreds of kilometres to seminars. Extension agencies in capital cities may lack funds and staff to update old guides. TechRadar cites a McKinsey survey reported by Bloomberg that about 17 percent of farmers worldwide already use generative AI in daily decisions.

TechRadar says this uptake is higher than for robotics or electric machinery. The McKinsey survey therefore places generative AI ahead of other new farm technologies.

What field-specific AI would need

TechRadar argues that the problem is not AI itself. The issue is that general-purpose chatbots, which are built for many tasks, are not trained for farming. They may know crop types and standard practices, but they do not ask for field details before giving advice.

A safer system would learn from satellite images, past harvests, weather and crop growth stage. It would also learn from working agronomists, or farming scientists, who can set limits on its answers. For example, a generic chatbot might suggest a strong chemical for wheat without knowing the crop is at a fragile stage or that high winds are forecast. A data-connected system could check weather and satellite images, choose a safer option and mark the exact areas to treat.

If the model knows the crop, it can avoid recommending a banned herbicide. It can also ask for more details when data is incomplete.

Specialised assistants and barriers

The article predicts a market for specialised agricultural AI assistants. These tools could connect to dashboards and satellite imagery, letting users move between data sources. Some can review satellite layers and answer questions about changes over the last 48 hours.

Others accept voice messages about treatments and store them as field records. TechRadar says this could reduce errors and save money. It also warns that asking farmers to enter all data into a generic chatbot is unrealistic. Many growers lack time and technical knowledge for regular data entry.

The aim is to make field records easier to keep current. That could help models give more useful answers.

Sources

  1. Why general-purpose AI falls short in agricultureTechRadar

Tags

Spotted a mistake? See our corrections policy. Read how News by OnDevice works in our editorial policy.